Abstract
A first‐order Markov chain and an alternating renewal process (ARP) with a truncated geometric distribution of wet day intervals and a truncated negative binomial distribution of dry day intervals are compared as models describing the occurrence of sequences of wet and dry days. Numerical optimization techniques are used to obtain approximate maximum likelihood estimates of the Fourier coefficients which describe the seasonal variation of the two Markov chain parameters and the three parameters in the alternating renewal process. For the four U.S. stations studied, the Markov chain model was superior to the ARP using the minimum Akaike information criterion.

This publication has 10 references indexed in Scilit: